Trust and compliance: evidence from the EU emissions trading scheme Ara Jo May 2019

Centre for Climate Change Economics and Policy Working Paper No. 333 ISSN 2515-5709 (Online) Grantham Research Institute on Climate Change and the Environment Working Paper No. 298 ISSN 2515-5717 (Online)

This working paper is intended to stimulate discussion within the research community and among users of research, and its content may have been submitted for publication in academic journals. It has been reviewed by at least one internal referee before publication. The views expressed in this paper represent those of the authors and do not necessarily represent those of the host institutions or funders.

The Centre for Climate Change Economics and Policy (CCCEP) was established by the University of Leeds and the London School of Economics and Political Science in 2008 to advance public and private action on climate change through innovative, rigorous research. The Centre is funded by the UK Economic and Social Research Council. Its third phase started in October 2018 with seven projects:

1. Low-carbon, climate-resilient cities 2. Sustainable infrastructure finance 3. Low-carbon industrial strategies in challenging contexts 4. Integrating climate and development policies for ‘climate compatible development’ 5. Competitiveness in the low-carbon economy 6. Incentives for behaviour change 7. Climate information for adaptation

More information about CCCEP is available at www.cccep.ac.uk The Grantham Research Institute on Climate Change and the Environment was established by the London School of Economics and Political Science in 2008 to bring together international expertise on economics, finance, geography, the environment, international development and political economy to create a world-leading centre for policy-relevant research and training. The Institute is funded by the Grantham Foundation for the Protection of the Environment and a number of other sources. It has six research themes:

1. Sustainable development 2. Finance, investment and insurance 3. Changing behaviours 4. Growth and innovation 5. Policy design and evaluation 6. Governance and legislation

More information about the Grantham Research Institute is available at www.lse.ac.uk/GranthamInstitute Suggested citation: Jo A (2018) Trust and compliance: evidence from the EU emissions trading scheme. Centre for Climate Change Economics and Policy Working Paper 333/Grantham Research Institute on Climate Change and the Environment Working Paper 298. London: London School of Economics and Political Science

Trust and Compliance: Evidence from the EUEmissions Trading Scheme

Ara Jo∗

Abstract

I study the role of trust in firms’ compliance decisions in the context of the EUEmissions Trading Scheme, an international regulation implemented in multiplecountries with different levels of trust. I find that trust in the country where theinstallation operates has a strong positive influence on its compliance decision.Including country fixed effects by exploiting the differences in the location ofheadquarters of multinational firms provides further evidence on the positiveimpact of trust on compliance.

Keywords: Trust, compliance, EU ETS.JEL Classification: Q54, Z1, K32.

∗Center of Economic Research, ETH Zurich, Zurichbergstrasse 18, 8089 Zurich, Switzerland.Email: ajo@ethz.ch. I am immensely grateful to Antoine Dechezlepretre and Steve Gibbons for theircontinuous support and guidance. I also thank Tim Besley, Stefano Carattini, Maureen Cropper,Maitreesh Ghatak, Bard Harstad, Vernon Henderson, Stefania Lovo, Daire McCoy, Antony Mill-ner, Francesco Nava, Henry Overman, Rick van der Ploeg, Sefi Roth, Olmo Silva, Thomas Stoerk,Ulrich Wagner and participants in various seminars at the LSE, Mines ParisTech, Oxford, FSR Cli-mate Annual Conference, and the Annual Meeting of the Spanish Economic Association (SAEe) fortheir helpful comments. All errors are my own. I gratefully acknowledge financial support from theGrantham Foundation for the Protection of the Environment through the Grantham Research In-stitute and the Marshall Institute for Philanthropy and Social Entrepreneurship, London School ofEconomics.

1 Introduction

The importance of generalized trust – the expectation that a random member of society

is trustworthy – in economic outcomes has gained recognition in the literature. In

particular, a number of papers have studied the influence of trust in the design and

stringency of regulations (Algan and Cahuc, 2009; Aghion et al., 2010; Aghion, Algan

and Cahuc, 2011). A related question is then whether and to what extent trust affects

compliance of regulated entities under a given regulation. Although the effect of formal

institutional measures on compliance is well-documented, there is a dearth of empirical

evidence on the effect of informal institution, such as the culture of generalized trust,

on compliance.1

To address the gap in the literature, I study how trust affects compliance decisions

in the context of the European Union Emissions Trading Scheme (EU ETS): the world’s

largest carbon trading market. This setting offers a number of advantages. First, it

provides an ideal environment in which the same legislation is implemented in multi-

ple countries, thus allowing me to investigate the systematic differences in compliance

behavior caused by cultural traits such as trust, which largely varies at the country

level.2 Relatedly, the penalty for noncompliance is set at the EU level. This feature

substantially reduces the problem of having differential levels of severity in punishment

for violation. Finally, the European Union Transaction Log (EUTL), a system harmo-

nized at the EU level, provides detailed installation level compliance data comparable

across countries. Existing papers that studied compliance of firms under environmental

regulations have used data on a single industry or several industries in a single country

(e.g. Gray and Deily, 1996; Shimshack and Ward, 2005, 2008; Dasgupta, Hettige and

Wheeler, 2000; Nyborg and Telle, 2006; Duflo et al., 2014; Evans, 2016). I address

this lacuna by taking advantage of this unique international dataset that contains over

16,000 installations operating in 31 different countries.

Identifying the role of trust in compliance is confounded by the task of having

1Luttmer and Singhal (2014) note that there is a lack of attention to social and intrinsic motivationsbehind tax compliance and only recently papers have started to empirically investigate the importanceof these nonpecuniary motivations (e.g. Dwenger et al., 2016; Hallsworth et al., 2017; Besley, Jensenand Persson, 2019). A similar gap exists in the literature on compliance in environmental regulation,to which the current paper is more closely related (Gray and Shimshack, 2011).

2Some papers have also exploited within-country variation in trust. For instance, Guiso, Sapienzaand Zingales (2004) study the effect of trust on financial development in Italy, a country known for itssubstantial cultural variation across regions. Tabellini (2010) also exploit regional variation in trustacross 8 large European countries.

1

to disentangle the effect of legal enforcement from the role of trust. Although the

stringency of formal enforcement is harmonized at the EU level, it is likely that country-

specific regulatory environment or institutional capacity is correlated with how the rules

are actually enforced in each country. Given that previous studies have documented

a strong influence of trust on the design of institutions and regulations (Algan and

Cahuc, 2009; Aghion et al., 2010; Aghion, Algan and Cahuc, 2011), it is then likely

that trust picks up the effect of formal enforcement on compliance, rather than trust

per se.

I attempt to circumvent this difficulty with two approaches. First, I instrument

the average level of trust in each country with trust inherited by second-generation

immigrants whose parents came from these countries. Given the persistence of trust

across generations (Rice and Feldman, 1997; Guiso, Sapienza and Zingales, 2006),

inherited trust observed in second-generation immigrants is expected to be correlated

with the level of trust in their countries of origin where their parents came from, and

yet unlikely to directly affect compliance behavior of firms operating in their source

countries since they are born and reside in their adopted countries. Second, based

on the documented influence of the source-country characteristics in multinational

enterprises (MNEs)’ operation abroad (Bloom and Van Reenen, 2007; Burstein, Monge-

Naranjo et al., 2009; Bloom, Sadun and Van Reenen, 2012b), I investigate if trust in the

country where the multinational is headquartered has influence on compliance decision

in the affiliate’s foreign location. This specification allows me to include country of

operation fixed effects, which removes any bias associated with unobservable national

characteristics that may be spuriously correlated with trust and compliance.

Consistent with the main prediction of the conceptual framework, I find that trust

prevalent in the country where the installation is located has a strong positive influence

on its compliance decision. Further, exploiting the differences in the location of global

headquarters of MNEs reveals that installations owned by firms headquartered in high-

trust countries are more likely to comply with the regulation than installations owned

by firms based in low-trust countries, even when they operate in the same geographic

area (country as well as region): for example, in Germany, an installation operated by

a multinational firm headquartered in Norway (a high-trust country) would be more

likely to be in compliance with the EU ETS than an installation owned by another firm

whose global headquarters are located in Greece (a low-trust country). The magnitude

of the estimated effect is economically meaningful: a change in ownership from a

2

multinational firm based in the lowest-trust country in my sample (Philippines) to

another MNE headquartered in the highest-trust country (Norway) would be associated

with a 1.5 percentage point decrease in the probability of noncompliance when the

average noncompliance rate is 3.2 percent. This effect is comparable to the previous

estimates for the effectiveness of traditional formal enforcement measures.

This paper contributes to the established literature on the effect of trust on vari-

ous economic outcomes. Trust has been shown to affect economic development (e.g.

Tabellini, 2010; Algan and Cahuc, 2010), financial development (Guiso, Sapienza and

Zingales, 2004), trade patterns (Guiso, Sapienza and Zingales, 2009), and global coop-

eration (Jo and Carattini, 2018). This paper presents the first empirical evidence on

the effect of trust on compliance and by doing so, provides a micro-empirical foundation

for a number of papers that study how trust affects the design of formal institutions

through the degree of law-abidingness of citizens in society (Algan and Cahuc, 2009;

Aghion et al., 2010; Aghion, Algan and Cahuc, 2011).

Also relevant is the literature that investigates firms’ compliance in environmental

regulations. A number of papers have documented strong deterrent effects of formal

enforcement actions (Gray and Deily, 1996; Deily and Gray, 2007; Shimshack and

Ward, 2005, 2008; Dasgupta, Hettige and Wheeler, 2000; Nyborg and Telle, 2006; Telle,

2013; Duflo et al., 2014; Evans, 2016). My paper differs from the existing literature

as I investigate the effect of informal institution – the culture of generalized trust –

on the enforcement of regulation, which has received relatively little attention in the

literature despite its documented importance and effectiveness (Ostrom, 1990, 2000).

The article is organized as follows. Section 2 presents a simple conceptual framework

along with supporting evidence. Section 3 provides background information on the

institutional setting and Section 4 describes data used for the analysis in detail. Section

5 presents the empirical analysis and Section 6 concludes.

2 Conceptual framework

In this section I discuss two main mechanisms behind the association between the

culture of trust and compliance based on the existing literature. I then focus on the

common prediction of both theories that higher levels of generalized trust positively

affect compliance.

3

2.1 Internalized norms

Sociologists have documented that individuals in high-trust societies tend to have

strong internalized norms: they donate to charity, obey traffic rules, and pay taxes

because they feel obligated to do so (Portes, 1998). In a similar spirit, trust shared in

an area may positively affect compliance in environmental regulation.

The growing literature on corporate culture suggests such internalized norms are

present at the firm level. Most firms have clearly defined corporate culture – principles

and values that should inform the behavior of all the firm’s employees (Guiso, Sapienza

and Zingales, 2015b).3 It is then likely that these self-declared values are influenced

by internalized norms of the individuals who form the organizations and those of the

region where they operate.

In parallel to the literature on the transmission of culture by immigrants that

emphasizes the strength and persistence of internalized norms (Fernandez and Fogli,

2006, 2009; Giuliano, 2007), there is a growing interest in the influence of source-country

characteristics in multinational firms’ operation abroad. For instance, Bloom, Sadun

and Van Reenen (2012b) provide empirical evidence that firms in high-trust areas

tend to be more decentralized since trust facilitates delegation of decision making

power from the CEO to managers. More interestingly, they find that trust in the

country where multinational firms are headquartered has a strong positive correlation

with decentralization in the affiliate’s foreign plants, even when the managers in the

plants are hired locally. This finding suggests that firms also take some of their home

country culture abroad and emphasizes the strong presence of internalized corporate

culture within firms. Thus, I expect trust in the country where firms operate to be

positively correlated with their compliance behavior and to potentially affect their

foreign subsidiaries’ compliance decisions through shared corporate culture.

2.2 Social punishment for noncompliance

Alternatively, assuming that social sanctions for noncompliance are stronger when

compliance rate is higher, trust may affect compliance decisions of firms through a

high expected compliance rate in society. In reality, the ‘name-and-shame’ sanction in

the EU ETS whereby member states “ensure publication of the names of operators and

3Guiso, Sapienza and Zingales (2015b) report that when they looked at companies’ web pages,they found that 85% of the Standard and Poor’s 500 (S&P 500) companies had a section dedicatedto corporate culture.

4

aircraft operators who are in breach of requirements to surrender sufficient allowances”

(Article 16(2) of the Directive 2003/87/EC, henceforth the Directive), clearly embodies

the threat of social punishment for noncompliance.4

I provide a simple analytical model of firms’ pollution behavior in the presence of

regulation to formalize this reasoning.5 Let firm i with emission intensity (or simply

type) θi ∈ [0, 1] choose an action bi ∈ [0, 1] when there is a regulation L ∈ [0, 1] that

imposes an upper bound on firms’ actions.6 The payoff is represented by:

ui(bi, θi) = −a(bi − θi)2 − (1− a)(bi −Bj)2 − γ

∑i∈N

bi − µI{bi>L}φ (1)

where the element of trust shared in society j is introduced by Bj as the expected

action of other firms. I assume that higher trust leads to a lower expected polluting

action of other firms. The parameter a ∈ (0, 1) is an (inverse) measure of social sensi-

tivity that governs the relative importance of matching one’s own type versus matching

the expected average action of other firms as shown in the first two terms.7 γ captures

negative externalities from the total emissions in society. It justifies interpreting lower

Bj as higher trust because given this society-wide externality, trust or perceived trust-

worthiness of other firms leads to the expectation that other firms will choose lower

actions for the common good. The last term subtracts the fine φ imposed on noncom-

pliant firms, those choosing bi > L, conditional on there being a formal inspection by

the authority with probability µ.

Given the set-up, there is a threshold type θ∗ = f(Bj, L, µ, φ), beyond which all

types violate the regulation and below which all types comply with the regulation.8 It is

then straightforward to show that fewer firms violate the regulation (i.e., θ∗ increases)

when trust is higher (Bj is lower), the regulation is less stringent (the upper bound L

4Banerjee and Shogren (2010) and Qin and Shogren (2015) provide theoretical frameworks forsocial motivations behind firms’ compliance in environmental regulation.

5The set-up is adopted from Acemoglu and Jackson (2017) where they study the role of socialnorms in the enforcement of laws.

6Firm i’ type θi ∈ [0, 1] is distributed according to a cumulative distribution function F . Forsimplicity, F is assumed to be strictly increasing and continuous on [0,1] with F (0) = 0 and F (1) = 1.

7Note that deviation in both directions is equally costly. For deviation from the prevailing action inthe opposite direction (complying when others violate), Fehr and Gachter (2000) provide experimentalevidence that there is a strong aversion against being the “sucker” who cooperates when others donot.

8The existence of a threshold type follows from the monotonicity of the first order conditions. Iprovide a formal proof of this statement in the Online Appendix.

5

is higher), and formal enforcement rules are stricter (µ and φ higher).9

2.3 Supporting evidence

The two potential channels discussed so far share the prediction that higher levels of

trust lead to more compliance. Here I provide descriptive evidence that is consistent

with the prediction. Figure 1 illustrates that there is indeed a negative correlation

between trust and noncompliance rates in the EU ETS across countries. While some

countries are close to full compliance, other countries such as Bulgaria, Italy and Slo-

vakia (relatively low-trust countries according to several international social surveys)

display very high noncompliance, over or close to 10 percent.

[Figure 1]

Micro-level regressions in Table 1 confirm the negative correlation between trust and

noncompliance. I begin by regressing the binary noncompliance variable that takes 1

if the installation is noncompliant and 0 otherwise against the trust measure of the

country where the installation is located, without any controls (column (1) of Table

1).10 In the next columns, I include an increasingly extensive set of controls that may

affect compliance at the country level and at the firm level. The correlation between

trust and noncompliance, however, seems to exist independently over and above these

factors.

[Table 1]

Although the common prediction is supported, it is challenging to distinguish the

two potential channels empirically. Later I attempt to provide suggestive evidence

for the presence of internalized corporate culture by looking at multinational firms’

compliance behavior when they operate abroad.

9Similarly, I formally show ∂θ∗

∂Bj< 0 in the Online Appendix.

10The variables used here will be described in detail in Section 4. The set of controls in the nextcolumns includes year and industry fixed effects, country-level countrols (rule of law, perceived regu-latory quality, log GDP per capita, percentage of population with tertiary education, log population)and firm-level controls (the number of employees, revenue, total assets and the number of installa-tions).

6

3 Institutional background

Launched in 2005, the European Union Emissions Trading Scheme (EU ETS) is the

world’s first and the largest carbon trading market operating in 31 countries (all 28

EU countries plus Iceland, Liechtenstein and Norway). It limits emissions from heavy

energy-using installations (including power stations and industrial plants) and airlines

operating between these countries covering around 45% of the EU’s greenhouse gas

emissions. Its geographic coverage, as large as all of Europe, offers a unique setting

to investigate the extent to which compliance behavior with respect to the same reg-

ulation may differ across countries due to the differences in trust and civicness of the

population.

According to the annual compliance cycle, operators of industrial installations and

aircraft operators (henceforth called installations) report their emissions of the previous

year verified by third-party accredited verifiers by 31 March of each year. Installations

are then required to surrender a quantity of allowances equal to the volume of their

verified greenhouse gas emissions of the previous year by 30 April of that year. An

installation is considered out of compliance if the number of allowances surrendered by

30 April is lower than its verified emissions.11 Noncompliant installations are subject

to the EU level penalty for the amount of emissions for which the installation failed

to surrender allowances (40 euro per tCO2 in phase 1 and 100 euro per tCO2 in phase

2 and 3) and the shortfall in compliance is then added to the compliance target of

the following year (i.e. paying a fine does not exempt noncompliant installations from

their obligations to surrender sufficient allowances).

To ensure compliance, each national government is required to lay down penalties

that are “effective, proportionate, and dissuasive” (Article 16(1) of the Directive) and

to submit a report on the implementation of the EU ETS to the European Commission

that contains information on monitoring and enforcement activities of the regulatory

body running the regulation. I analyzed these reports and found that enforcement

was generally weak across countries. Table A1 shows whether there were any penalties

imposed on violators in each country in phase 1 when noncompliance was considerably

higher than in the following phases. For instance, of the countries that submitted

a report for 2005 (four countries did not) only three (Portugal, Spain and the UK)

11There are other forms of noncompliance such as failing to report changes in the installation’scapacity or monitoring plans. However, data on these forms of noncompliance comparable acrosscountries are unavailable.

7

issued penalties on violators although most countries had violators in that year. A

policy report prepared by the European Commission (European Court of Auditors,

2015) confirms that most countries are not successful in implementing EU ETS-related

penalties. European Court of Auditors (2015) explains that countries are often limited

in their own legal and administrative capacity for the successful implementation of EU

ETS penalties. Institutions running the regulation are either not empowered to impose

sanctions themselves (e.g. Italy) or need to await the outcome of lengthy court pro-

cedures and appeals (e.g. Germany).12 The report adds that on-the-spot inspections

to assess the implementation of the self-monitoring plan submitted by installations

were also very limited.13 This lack of strong institutional enforcement may explain

the existence of noncompliance despite the fact that the cost of purchasing allowances

was well below the penalty for not surrendering sufficient allowances throughout the

sample period (Figure A1). It also makes the EU ETS a suitable context to study the

influence of culture on compliance, which would be minimal in the presence of perfect

monitoring and enforcement.

Although enforcement was weak generally across countries, the presence of these

country-specific enforcement rules introduces difficulties in identifying the role of trust

in compliance (i.e. it would be problematic if high-trust countries also have more strin-

gent enforcement rules and more frequent inspections).14 In later sections, I propose

identification strategies that overcome this obstacle.

4 Data description

4.1 Compliance in the EU ETS

Data on compliance is provided by the European Union Transaction Log (EUTL), a

system harmonized at the EU level that publishes information on compliance status,

12Indeed, the German authority mentions in their 2007 report that penalties for violators in 2005were issued only in 2007 and a majority of the cases (11 out of 16) are on appeal.

13European Court of Auditors (2015) explains that other types of visits to installations were of-ten performed in the context of the Integrated Pollution Prevention and Control (IPPC, Directive2008/1/EC) or other environmental legislation that were considered to be of higher priority (e.g. inFrance, Germany and Poland) without specifically addressing EU ETS-related issues.

14A related concern raised by Ellerman and Buchner (2007) is the possibility that some memberstates systematically overallocated permits to their installations in phase 1 and 2 when allocation wascarried out at the national level (from phase 3 allocation is done at the EU level). To the extent thatthe degree of overallocation might be correlated with trust this could also bias the estimate for theeffect of trust on compliance.

8

permit allocation, verified emissions, and surrendered allowances at the installation

level. Existing papers that have studied compliance behavior of firms have focused on

a single industry or several industries in a single country.15 While providing valuable

insights into various motivations behind compliance decisions, these studies are unable

to shed light on the systematic differences in compliance behavior caused by cultural

traits such as trust, which largely varies at the country level. I address this lacuna

by taking advantage of this unique international dataset that contains installations

operating in multiple industries and multiple countries.16

I use information on compliance status from 2005 to 2015 that includes all three

phases so far. There are five possible compliance codes installations can be given: (1)

A, when the number of allowances and permits surrendered by the deadline (30 April)

is greater than or equal to verified emissions, (2) B, when the number of allowances

and permits surrendered by the deadline is lower than verified emissions, (3) C, when

verified emissions were not entered until the deadline, (4) D, when competent authority

corrected verified emissions after the deadline and decided that the installation is not in

compliance, and (5) E, when competent authority corrected verified emissions after the

deadline and decided that the installation is in compliance. The distribution is reported

in Table A2 in the Appendix in detail. Based on this categorization, I construct a binary

noncompliance variable that takes 1 if an installation is given either B or D and 0 if

an installation is given either A or E. In my preferred specification, I treat compliance

status of code C as missing in order to be conservative.17 Alternative specifications

such as considering A and B only or treating C differently yield similar results.

[Figure 2]

15For single industry studies, see, for example, Gray and Deily (1996) for the US steel industry,Shimshack and Ward (2005, 2008) for the US pulp industry. Multi-industry studies include Dasgupta,Hettige and Wheeler (2000) for Mexico, Decker and Pope (2005) and Evans (2016) for the US, Nyborgand Telle (2006) for Norway, and more recently Duflo et al. (2014) for India.

16I drop Cyprus, Iceland, Liechtenstein, Malta, and Luxembourg since there are too few installations(less then 50) operating in these countries, thus may not represent the culture of the environmentin which they operate. Due to the small number of regulated installations, in some cases, technicalaspects of monitoring, reporting and verification procedures were delayed (for instance, Malta), whichcould affect the compatibility of the data. I also drop Croatia that joined the EU ETS in 2013.

17Although failing to report verified emissions is strictly speaking noncompliance, two observationscall for a more cautious approach. First, among observations with compliance status C, around 80percent have incomplete information on permit allocation, either missing or zero even in the firsttwo phases when most permits are given for free based on their historical emissions. Second, theseinstallations tend to have missing verified emissions for multiple periods followed by missing compliancestatus in the following periods. Taken together, it is plausible that these installations were no longerregulated (or active) and therefore did not have reporting obligations.

9

The cross-country compliance rates depicted in Figure 2 reveals startling variation

across countries. It is noteworthy that the distribution is highly right-skewed with

a majority of countries close to full compliance and several countries with very high

noncompliance. Some countries such as Bulgaria, Italy, and Slovakia, have close to

or over 10 percent noncompliant installation-year observations. However, the mean

noncompliance rate is very low – 3.2 percent – and half the countries show less then 1

percent noncompliant observations during the sample period.18

4.2 Measuring trust

I build trust measures using the European Social Survey (ESS). I pool data from the

seven waves (from 2002 to 2014), which includes all European countries that participate

in the EU ETS. The ESS measures generalized trust – the expectation that a random

member of the society is trustworthy – by asking the classical question, “Generally

speaking, would you say that most people can be trusted, or that you can’t be too

careful in dealing with people?”.19 Respondents’ answers are given on a scale of 0

to 10, where 0 implies “You can’t be too careful” and 10 means “Most people can

be trusted”. The variable that I use in the econometric regression is the average of

this answer within the country where the installation is located. For identification

purposes, I later also explore the importance of trust prevalent in the location of

the firm’s headquarters (that owns the installation) when the firm’s headquarters are

located in a different country and therefore likely to be exposed to a different set of

values and corporate culture.

This trust question appears in several other surveys including the World Value

18The occurrence of noncompliance was very high in 2005 (59% of all noncompliance occurring in thefirst year) and the rate was substantially lower in 2006 onwards (Figure A2 in the Appendix). In casethe cross-country pattern observed in 2005 is an outlier I exclude 2005 and calculate noncompliancerates across countries. A similar cross-country pattern continues to exist, albeit with lower magnitudes,as shown in Figure OA1 in the Online Appendix with Bulgaria, Italy and France appearing high inthe ranking. Even without 2005, noncompliance tends to be less frequent in phase 2 and phase 3. Onepossible reason is that the EU level fine for the amount of emissions for which the installation failedto surrender allowances increased by 2.5 times starting from phase 2 (40 euro per tCO2 in phase 1 to100 euro per tCO2 in phase 2 and 3).

19The ESS also measures trust that respondents have in parliament, legal system, the police, politi-cians and political parties. These measures can also explain compliance patterns and are highlycorrelated with the measure of generalized trust (the smallest pairwise correlation coefficient being0.84 between trust in parliament and generalized trust). However, they are likely to reflect the qualityof the corresponding institutions, whose effect on compliance I try to remove in order to focus on theeffect of trust as culture. Thus I believe it is appropriate to focus on this measure of generalized trustin my analysis.

10

Survey (WVS) with the same wording and has been the most widely used tool to

measure trust across countries in the literature. A number of papers have confirmed

that it is indeed correlated with trusting behavior. Fehr et al. (2003) show that survey

questions of this type do capture trust by running a series of experiments, and Fehr

(2009) further demonstrates that the survey measure of trust is strongly correlated

with the behavioral measure of trust derived from trust games. On the contrary,

Glaeser et al. (2000) provide experimental evidence that the survey question captures

the trustworthiness of respondents rather than trust; but, this conflicting finding has

been reconciled by Sapienza, Toldra-Simats and Zingales (2013) who show that subjects

in a homogeneous sample (such as Harvard undergraduates as in Glaeser et al. (2000))

tend to extrapolate the trustworthiness of others based on their own trustworthiness.

However, in a large anonymous sample (such as random individuals in Germany as in

Fehr et al. (2003)) in which respondents are not extrapolating expected trustworthiness

of others based own their own trustworthiness, the survey question does seem to capture

trust. Thus, I believe the trust measure from the ESS is appropriate for the purpose

of my analysis that investigates the role of trust in compliance decisions by firms.

[Figure 3]

Figure 3 plots the average level of trust by country. Two points are noteworthy.

First, as shown in previous studies, there exists substantial variation in trust across

countries. The average level of trust ranges from a minimum of 3.8 observed in Portugal

to a maximum of 6.9 in Denmark. Second, it is readily observable that there are

differences across regions of Europe; for instance, Nordic countries (Denmark, Norway,

Finland, and Sweden) display highest levels of trust in the sample. On the other hand,

Mediterranean countries such as Greece, Italy, and Portugal appear to have lower levels

of trust. Continental European countries tend to be in the middle of the trust ranking.

4.3 Firm-level controls

Data on firm characteristics comes from Bureau Van Dijk’s Orbis Database. The

account holders’ information in the EU ETS (i.e., regulated installations) was matched

to the corporations in the Orbis Database in Calel and Dechezlepretre (2016). I obtain

key financial variables that may affect compliance decisions (i.e., firms may be too

financially constrained to buy enough permits) including the number of employees,

operating revenue and total assets for the sample period as well as firms’ ownership

11

structure in 2015 and the number of installations run by each firm. These controls

will also account for firm-level heterogeneity more generally. Table OA1 in the Online

Appendix reports the descriptive statistics of these variables for firms in each country.

5 Trust and compliance

The discussions in the conceptual framework in Section 2 predict that greater trust

leads to higher compliance, or fewer firms violating the regulation. In this section, I

subject this prediction to rigorous econometric investigation.

5.1 Using inherited trust as instruments

The negative correlation between trust and noncompliance documented in Section 2 is

consistent with the theoretical prediction. However, it is possible that trust picks up the

effect of country-specific regulatory environment or institutional capacity that might be

correlated with trust, given the documented influence of trust in shaping institutions

and regulations (Algan and Cahuc, 2009; Aghion et al., 2010; Aghion, Algan and Cahuc,

2011). Relatedly, Cohen (1998) and Brehm and Hamilton (1996) have argued that the

presence and characteristics of other environmental regulations may affect compliance

behavior of firms through higher degrees of familiarity and knowledge with compliance

procedures. I included a measure that controls for rule of law and the perceived quality

of regulation in section 2.3, but it may not be perfect.

It is also plausible that the correlation could also be explained by some unobservable

factors that affect regulatory compliance of firms and the level of trust within the

country simultaneously.20 For instance, Jo and Carattini (2018) document that high-

trust countries have reduced their per capita CO2 emissions more substantially than

low-trust countries between 1950 and 2010. Then, one might argue that it might

be easier for installations in high-trust countries to comply with the EU ETS since

they already operate in an environment more conducive to reducing emissions. Thus,

what I need is a measure that can predict the average level of trust in a country, but

uncorrelated with country-specific formal institutions and other unobservable features

that may affect compliance behavior of firms.

20I believe the threat of reverse causality is minimal given the extensive evidence on the importanceof historical determinants of trust (Guiso, Sapienza and Zingales, 2009; Tabellini, 2010).

12

One such measure is the inherited component of trust observed in second-generation

immigrants. This epidemiological approach has gained recognition in the literature

(Fernandez, 2007) and been adopted by several papers that attempt to isolate the

causal effects of trust on economic outcomes (Algan and Cahuc, 2010; Butler, Giuliano

and Guiso, 2016; Jo and Carattini, 2018). The insight is based on the evidence that

trust is highly persistent across generations through the transmission of values within

families (Rice and Feldman, 1997; Guiso, Sapienza and Zingales, 2009). Then, inherited

trust observed in second-generation immigrants is expected to be correlated with the

level of trust in their countries of origin where their parents came from, and yet unlikely

to directly affect compliance behavior of firms operating in their source country since

they are born and reside in their adopted countries.

I apply this idea to my analysis by using, for example, the average level of trust

among second-generation British immigrants born and raised in any of the other ESS

countries to predict the level of trust in Britain. The exclusion restriction is then trust

of second-generation British immigrants born and living in Spain, for instance, should

not directly affect compliance decisions of regulated firms operating in Britain between

2005 and 2015. The number of second-generation immigrants from each country from

which I estimate this measure of inherited trust is reported in Table OA2 in the Online

Appendix. Figure 4 clearly depicts a strong positive correlation between the inherited

trust of immigrants and the level of trust observed in their source county, which ensures

a strong first stage.

[Figure 4]

I estimate regression equations of the following form:

Noncomplianceijct = α + βTrustc + φCct + ρFijct + δY eart + ξIndustryj + εijct (2)

where Noncomplianceijct is a binary variable that takes 1 if firm i in industry j in

country c is out of compliance in year t. Trustc is the average trust of country c where

installations are located. It is reasonable to suppose that the variable does not vary

over time during the 11-year period I study, given the persistent nature of trust across

generations.21 Most empirical analyses in the trust literature follow this approach by

21To formally test if there is time variation over the study period I check whether there is overlap inthe 90% confidence intervals of the trust variable for the start and end year using 2000 and 2014 wave,respectively. Only two out of 25 countries in my sample have non-overlapping confidence intervalsover this period.

13

taking the average of trust in surveys conducted since the 80s (e.g. Tabellini, 2010;

Bloom, Sadun and Van Reenen, 2012b).22 Therefore, I run a pooled regression despite

the panel nature of the dependent variable. To avoid understating the standard errors

due to repeated observations, the errors are clustered at the country level over all years.

Cct and Fijct represent country-level controls and firm-level controls. I further include

year dummies and industry dummies.

Table 2 reports IV probit estimates. Column (1) first shows the IV estimates from

the regression that does not include any controls to begin with. The coefficient on trust

is negative and statistically significant (with P-value 0.018). The instrument is strong

with F -statistics over 40. Column (2) includes year dummies and industry dummies,

which will capture industry-specific characteristics that may affect compliance such as

available abatement technology or market situations. In column (3), I include country-

level controls such as log GDP per capita, log population, educational attainment

and two governance indicators that measure country-wide rule of law and perceived

regulatory quality (summary statistics of these variables are reported in the Online

Appendix). In column (4), I further include firm-level variables such as the number

of installations each firm owns (to control for economies of scale in compliance) and

operating revenue, total assets and number of employees to control for the possibility

that firms were too financially constrained to buy permits. Due to the large number of

missing values in these firm-level financial variables, the sample size falls substantially,

and yet the negative relationship between trust and noncompliance remains robust.23

Column (5) shows the reduced form relationship between inherited trust and noncom-

pliance. The 2SLS estimate from a linear probability model is qualitatively similar

with a coefficient (standard error) of -0.049 (0.028).

[Table 2]

22Few studies exploit time variation in trust with a notable exception being Algan and Cahuc(2010). They suggest a methodology to recover long intertemporal variation in trust by comparingimmigrants who moved to America from different countries at different points in time and generatea trust measure for 25 countries with time variation over 60 years, which arguably covers multiplegenerations. Their trust variable measures trust in two points far apart in time, 1935 and 2000, toallow sufficient time for the evolution of trust. Algan and Cahuc (2009) also exploit time variationin trust over 20 years in one of the specifications, using only the end points of their data (1980 and2000) to get enough variation.

23To make sure the presence of missing values does not alter the distribution of compliance, I checkif the compliance rate differs with and without observations with missing firm-level controls and findthat the distribution of the dependent variable (noncompliance) is not statistically different across thetwo groups (with P -value of the test statistics 0.64).

14

The magnitude of the association between trust and compliance is substantial. The

estimate from column (4) that includes the full set of controls implies a 1 standard

deviation increase in trust (roughly from trust in Italy to trust in Netherlands) is

associated with a 2.4 percentage point decrease in the probability of noncompliance.

Some papers have exploited within-country regional variation in trust for identifica-

tion purposes (Guiso, Sapienza and Zingales, 2004; Tabellini, 2010; Bloom, Sadun and

Van Reenen, 2012b). I also try a trust measure at the region level in column (6) and

find consistent results. The most prominent benefit of exploiting regional variation in

trust is that it allows including country fixed effects and the studies mentioned above

successfully combine country fixed effects with region-level instruments to estimate the

causal effects of trust. However, the difficulty of adopting this approach in my context

is that country fixed effects will make it impossible to use my instrument, which is at

the country level (the region of origin of immigrants is not asked in the ESS).24 In the

next section, I suggest an alternative design that allows both using the instrument and

country fixed effects.

5.2 Exploiting differences in the location of headquarters

About 80 percent of installations (10,692 in total) for which I have ownership data are

owned by multinational firms (MNEs) and 4,310 of them are owned by foreign MNEs

whose central headquarters are located in a different country from the country where

the installations operate. This subsample offers a chance to further probe the causality

of the relationship that I attempted to estimate so far by allowing country of operation

fixed effects. Country fixed effects remove any bias associated with unobservable na-

tional characteristics that may be spuriously correlated with trust and compliance. I

then compare compliance behavior of installations that are exposed to the exact same

external environment (e.g. formal enforcement, stringency of other related regulations,

etc.) but have different levels of trust coming from the country of origin.

The importance of country of origin characteristics in MNEs’ management and or-

ganizational structure has long been recognized in the relevant literature. A study most

relevant to my analysis is Bloom, Sadun and Van Reenen (2012b) where they provide

24Another concern is that there might not be enough regional variation in trust once countrydummies are included, given that my sample includes 25 European countries, many of which areculturally homogeneous. Previous studies that exploited regional variation in trust focused on largecountries known for substantial within-country variation in trust including Italy (Guiso, Sapienza andZingales, 2004; Tabellini, 2010) or on the entire world Bloom, Sadun and Van Reenen (2012b).

15

evidence that the level of trust prevalent in the country where the multinational is

headquartered has a strong positive effect on the degree of decentralization in the affil-

iate’s foreign location (for instance, a Swedish affiliate operating in the US is typically

more decentralized than a French affiliate in the US). Furthermore, Bloom, Sadun and

Van Reenen (2012a) show that US multinationals operating in Europe displayed higher

productivity in the use of information technologies (IT) than non-US multinationals

in Europe during the period when the US experienced a rapid productivity growth

in sectors that intensively use IT. Burstein, Monge-Naranjo et al. (2009) and Bloom

and Van Reenen (2007) also document the transmission of knowledge and manage-

ment practices across countries in MNEs. Given this ample evidence on the influence

of source-country characteristics over MNEs’ operation abroad, it seems legitimate to

investigate whether there might be different patterns in compliance behavior across

multinationals based in different countries.25

The results of this analysis are reported in Table 3. For this exercise, I construct

another trust measure from the World Value Survey (WVS) since there are a number

of non-European countries in which MNEs in my sample are headquartered and thus

not included in the ESS.26 As before, I pool together individual responses from all six

waves conducted so far (1984, 1993, 1999, 2004, 2009, and 2014), and compute the

average level of trust in the country where the global headquarters of the installation

is located.27 I later also check for the independent role of trust in the installation’s

location.

[Table 3]

Column (1) shows the relationship between compliance and the level of trust in the

country where the central headquarters are located without any controls. Standard

errors are clustered at the country level. The coefficient is negative and significant at 1

25It is possible that some firms might have experienced changes in ownership through mergers andacquisitions (M&As) just before or while being subject to the EU ETS. To reduce the potential mea-surement error arising from this scenario, I identify firms that were bought out by foreign companies(i.e., target firms in foreign M&A deals) since 2000 using rich M&A data from Bureau Van Dijk’sZephyr Database. There are only a small number of such firms in my sample (264 out of 8,156 firms).I drop 573 installations owned by these firms from the regression.

26There are 44 source countries in my sample and the median (mean) number of firms headquarteredin each source country is 28 (103).

27The only difference in the trust question in these two surveys is the scale used for the answer.While the ESS uses the scale of 10, the WVS provides a binary choice between 0 and 1 where 0 implies“You can’t be too careful” and 10 means “Most people can be trusted”.

16

percent level, which suggests that trust prevalent in source countries is positively corre-

lated with the affiliates’ compliance decisions. The influence of trust in the country of

headquarters remains strong even when I control for individual firm-level characteris-

tics in column (2). Next, I include country of operation fixed effects as well as year and

industry fixed effects. The magnitude of the coefficient falls sharply with an extensive

set of fixed effects, but the coefficient in column (3) is still negative and significant.

This implies that installations owned by firms based in high-trust countries are less

likely to violate the regulation than those owned by firms in low-trust countries, even

when they operate in the same institutional environment. In column (4), I add the level

of trust in the location where the installation operates (at the region level, since the

country-level measure will be omitted due to country fixed effects). The coefficient on

trust in the region of installation is insignificant, while the role of trust in the country

of headquarters remains negative and statistically significant with a similar magnitude

as in column (3). One might worry that time-varying omitted variable bias may still

confound the estimate. For instance, after the 2011 nuclear accident in Japan, Ger-

many decided to dramatically reduce their dependence on nuclear power plants while

increasing the share of renewable sources in producing electricity. This led to changes

in regulatory environment and energy prices that might have affected firms’ compliance

behavior under the EU ETS. To deal with this concern of time-varying country specific

confounders, I include country by year fixed effects in column (5). The sample size falls

as there are country-year pairs with perfect compliance, thus no variation; however,

the coefficient remains qualitatively similar.

A remaining concern is potential omitted variable bias related to trust in the coun-

try of headquarters. For instance, Bloom, Sadun and Van Reenen (2012b) show that

MNEs with headquarters in high-trust countries are larger in firm size and more pro-

ductive than those with headquarters in low-trust countries. If compliance behavior is

correlated with these firm characteristics related to trust that I cannot directly control

for (although I control for the number of employees), the estimate might be biased.

Thus, I apply the same instrument developed in the section above to further probe

the role of trust in compliance behavior. The measure of inherited trust observed in

second-generation immigrants is still valid in this context, since it predicts the level of

trust in their source countries but unlikely to be correlated with the organization and

performance (such as size and productivity) of MNEs headquartered in those coun-

tries. Column (6) reports the IV estimates. The coefficient is negative and of larger

17

magnitude when instrumented and still statistically significant. Column (7) shows the

presence of a negative and significant relationship between noncompliance and the mea-

sure of inherited trust in the reduced form. In column (8), I further include country (of

operation) by year fixed effects (as in column (5)). Even in this demanding specifica-

tion, the influence of trust in the MNE’s source country continues to exist. Repeating

this specification in a linear probability model yields similar results (unreported).28

Not only is the estimated effect of trust on compliance statistically significant, it

is also economically meaningful. The estimate in column (5) implies that a change

in ownership from a multinational firm based in Philippines (the lowest-trust country

in my sample) to another MNE headquartered in Norway (the highest-trust country)

would be associated with a 1.2 percentage point decrease in the probability of noncom-

pliance. How large is this effect relative to that of formal enforcement on compliance?

To provide a sense of magnitude, I compare this effect with other existing estimates for

the effectiveness of formal enforcement actions reported in previous papers. Estimates

for the effect of traditional regulatory measures (e.g. inspections and fines) range be-

tween 42 and 52 percent treatment effects (Gray and Shimshack, 2011).29 Also, Evans

(2016) documents that an information-based enforcement tool such as the “watch list”

in the Clean Air Act is associated with a 21 percentage point decrease in the prob-

ability of noncompliance, indicating a 29 percent treatment effect given the average

noncompliance rate 72 percent. Compared with these previous estimates, the effect of

trust still seems large: given the average compliance rate of 3.2 percent in my sample,

the predicted fall in the probability of noncompliance by 1.2 percentage point caused

by the change in ownership from a Filipino firm to a Norwegian firm implies a 37

percent treatment effect.

28It yields a coefficient (standard error) on the trust measure of -0.021 (0.008).29Deily and Gray (2007) studied the deterrent effects of regulatory measures on compliance in the

Clean Air Act using compliance data on large steel mills in the Unites States. They found that beingsubject to an enforcement activity in the prior two years decreased the probability of noncomplianceby 32 percentage point. Given the overall noncompliance rate 62 percent, the estimate suggets a 52percent treatment effect. In a similar context, using compliance data on pulp and paper mills Grayand Shadbegian (2005) found that a typical regulatory action decreased the probability of violationby 10 percentage point, implying a 42 percent treatment effect (with the average violation rate 24percent in the sample).

18

5.3 Discussions

The strong effect of trust in the country of headquarters on MNEs’ compliance provides

support for the channel of internalized corporate culture discussed in Section 2.1 In

the Appendix, I further explore this line of argument by investigating the influence of

bilateral trust between two countries on MNEs’ compliance. For instance, is a French

affiliate more likely to comply with the regulation in Belgium (that the French tend to

trust) than in the United Kingdom (that the French tend to distrust)? I find that the

effect of bilateral trust is negligible and firms operate in accordance with their source-

country culture regardless of where they operate (Table A3), which provides further

support for the presence of internalized corporate culture.

Yet, this evidence does not necessarily rule out the channel of social sanctions for

noncompliance. To the extent that MNEs care about their reputation back home (in

which case the expected action of neighboring firms will be that of firms operating in

their home countries), the channel of social punishment for violation may still be at play

(Wijen and van Tulder, 2011).30 This possibility makes it challenging to empirically

disentangle the two channels. The findings should therefore be taken as a reduced-form

average effect of these mechanisms.

Finally, the evidence presented in this paper is also consistent with the literature

that emphasizes the role of knowledge in compliance by distinguishing two sources of

noncompliance, namely, ignorance and evasion (Brehm and Hamilton, 1996; Cohen,

1998; Jørgensen and Pedersen, 2005). The strong effect of trust in the country of

MNEs’ headquarters on compliance might indicate that firms in high-trust countries

obtain information required for compliance more actively due to internalized corporate

culture or strong social sanctions for violation. However, once equipped with the set of

knowledge necessary to comply with the regulation firms may comply wherever they

operate, regardless of the strength of formal enforcement activities in the country of

operation since the cost of doing so is low.

30A recent example would be the Volkswagen diesel emissions scandal. When it broke on Friday18 September 2015 in the US, on the following Monday morning (before it became an internationalscandal) the company’s share price dropped almost 20 percent in the Frankfurt stock exchange and theGerman government issued a statement that condemns the company’s alleged violation of the US law.For a timeline of the scandal, see https://www.theguardian.com/business/2015/dec/10/volkswagen-emissions-scandal-timeline-events.

19

5.4 Robustness checks

In this section I report the results from a number of robustness checks. Table OA3

in the Online Appendix reports robustness checks for the cross-country analysis using

all firms (as in Section 5.1) where I try to (1) drop late joiners in the EU ETS, (2)

use alternative specifications for noncompliance, (3) use alternative measures of trust,

(4) use an alternative specification for the measure of inherited trust, and (5) check

if installations’ compliance behavior is different also at the intensive margin, i.e., if

the amount by which installations are noncompliant can also be explained by the level

of trust. Here I focus on the main results from the specification using MNEs that

includes country of operation fixed effects. First, I add region-level economic controls

(log GDP per capita, log population and the percentage of population with tertiary

education) in addition to country of operation fixed effects (column (1) in Table 4).

Also, I exclude Bulgaria and Romania that joined the EU ETS later, in case there

might have been technical difficulties arising from immature infrastructure. Bulgaria

and Romania started to participate in the ETS in 2007 when they joined the European

Union in the same year.31 Excluding these late joiners does not affect the relationship

between trust and compliance (column (2)).

[Table 4]

Next, I try alternative specifications for the binary noncompliance variable. In my

preferred specification, I dropped installations with compliance status C that did not

report their verified emissions (the step before they surrender corresponding amount

of permits) in order to be conservative because there is suggestive evidence that these

installations are no longer regulated or active (see footnote 17). Alternatively, I treat

these installations as noncompliant when they can be reasonably presumed to be active

by two standards: first, when they have non-missing information on permit allocation

in the current period and second, when they have non-missing compliance status other

than C in the following period. The regression in column (3) uses this alternative

measure of noncompliance. The magnitude of the coefficient on the trust measure falls

but it remains significant at 10 percent level. I also try to drop installations whose

verified emissions were corrected later by the competent authority (i.e. those with code

D and E) and find similar results (column (4)).

31Croatia also joined the ETS in 2013 and is already dropped from my sample along with five smallcountries with less then 50 installations.

20

In column (5) and (6) I try alternative measures of trust to get a sense of potential

measurement error in the trust variable. First, I construct a measure that takes into

account year-specific shocks since I pool multiple waves conducted in different years

to calculate the average level of trust in each country. Following Bertrand, Duflo and

Mullainathan (2004) and Guiso, Sapienza and Zingales (2009), I regress trust on year

dummies, form residuals, and then compute the means of these residuals by country.

Column (5) shows that the coefficient on this alternative measure of trust is still neg-

ative and significant at 1 percent level. Next, I try a potentially more demanding

approach that further takes into account individual respondents’ characteristics such

as gender, age, education and income as well as year dummies. To proceed, I follow

Algan and Cahuc (2010) by regressing trust on a set of individual characteristics, year

fixed effects and country fixed effects. The coefficients on the country fixed effects then

measure the average level of trust relative to the omitted reference country (Germany,

in this case). The results from the individual-level regression are reported in Table

OA5 in the Online Appendix. The fact that coefficients on the country dummies are

significant (with standard errors clustered at the country level) even after controlling

for a set of individual characteristics and year dummies confirm the substantial cross-

country variation in trust documented in the literature. Column (6) reports the IV

probit estimate from the specification using this alternative trust measure and shows

the results consistent with previous findings. The effect of trust on compliance remains

robust across different measures of trust.

Finally, I try an alternative specification of the instrument that imposes a minimum

of 25-year lag between the launch of the EU ETS and the year of immigration of the

second-generation immigrants’ parents as in Algan and Cahuc (2010). This is to further

ensure that the exogeneity assumption is satisfied, since I only use second-generation

immigrants born before 1980 and therefore whose parents must have left their source

countries before 1980, at least 25 years (one generation) prior to the start of the EU

ETS in 2005. Then it is even less likely that the level of trust transmitted by those who

left the country at least 25 years ago still affects compliance behavior of firms in that

country. Column (7) shows that the result with this instrument is very similar and the

modified instrument still has a strong predictive power (F -stat is 18.6). Column (8)

shows the presence of a negative and significant relationship between noncompliance

and the alternative instrument in the reduced-form regression.

21

6 Conclusion

In this article, I attempt to provide rigorous empirical evidence on the effect of infor-

mal institution, the culture of trust, on compliance. I find strong evidence that trust

positively affects compliance and more importantly, there exist systematic differences

in firms’ compliance patterns depending on the country in which they are headquar-

tered, even when they operate in the same geographic area with the same external

environment.

One interesting implication is related to the idea of using corporations as a lab in

which to study the role of culture. Although the role of culture in economic activi-

ties has long been recognized, economists’ attempts to develop a deeper insight into

specific workings of culture have not been straightforward because (1) it is difficult to

know where culture comes from, (2) it is sticky with rare drastic changes, and (3) even

when these cultural changes occur they take place over a long period with many other

things happening at the same time. Guiso, Sapienza and Zingales (2015a) note this

problem and suggest corporations as an alternative environment to study the role of

culture. This is indeed promising since with corporate culture, we know (1) when, how,

and based on what values corporations are founded, (2) corporate culture is subject to

more frequent changes (e.g., through hiring, firing and M&As), and (3) performance

is more easily measured (Guiso, Sapienza and Zingales, 2015a). There is an increas-

ing interest in this line of reasoning that sheds light on specific mechanisms behind

the documented effect of culture at the macro-level. For instance, Bloom, Sadun and

Van Reenen (2012b) provides evidence on the influence of trust in firms’ decision to de-

centralize, which allows more efficient resource allocation within and across firms that

leads to higher firm productivity and economic growth. This serves as microevidence

for the long-held belief that trust facilitates economic growth through lower transac-

tions costs (Arrow, 1972). Similarly, this article provides microevidence on the role of

trust in compliance and by doing so, validates the documented effect of trust on the

design of formal regulation through how law-abiding people are (Tabellini, 2008). I

concur with Guiso, Sapienza and Zingales (2015a) that these approaches substantially

enhance our understanding of how cultural norms affect economic behavior and relate

to formal institutions.

22

References

Acemoglu, Daron, and Matthew O Jackson. 2017. “Social norms and the en-

forcement of laws.” Journal of the European Economic Association, 15(2): 245–295.

Aghion, Philippe, Yann Algan, and Pierre Cahuc. 2011. “Civil society and the

state: The interplay between cooperation and minimum wage regulation.” Journal

of the European Economic Association, 9(1): 3–42.

Aghion, Philippe, Yann Algan, Pierre Cahuc, and Andrei Schleifer. 2010.

“Regulation and Distrust.” The Quarterly Journal of Economics, 125(3): 1015–1049.

Algan, Yann, and Pierre Cahuc. 2009. “Civic virtue and labor market institutions.”

American Economic Journal: Macroeconomics, 1(1): 111–145.

Algan, Yann, and Pierre Cahuc. 2010. “Inherited trust and growth.” The American

Economic Review, 100(5): 2060–2092.

Arrow, Kenneth J. 1972. “Gifts and exchanges.” Philosophy & Public Affairs, 343–

362.

Banerjee, Prasenjit, and Jason F Shogren. 2010. “Regulation, reputation, and

environmental risk.” Economics Letters, 106(1): 45–47.

Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. “How

Much Should We Trust Differences-in-Differences Estimates?” The Quarterly Jour-

nal of Economics, 249–275.

Besley, Timothy, Anders Jensen, and Torsten Persson. 2019. “Norms, enforce-

ment, and tax evasion.” National Bureau of Economic Research.

Bloom, Nicholas, and John Van Reenen. 2007. “Measuring and Explaining Man-

agement Practices Across Firms and Countries.” The Quarterly Journal of Eco-

nomics, 122(4): 1351–1408.

Bloom, Nicholas, Raffaella Sadun, and John Van Reenen. 2012a. “Americans

do IT better: US multinationals and the productivity miracle.” The American Eco-

nomic Review, 102(1): 167–201.

23

Bloom, Nicholas, Raffaella Sadun, and John Van Reenen. 2012b. “The

Organization of Firms Across Countries.” The Quarterly Journal of Economics,

127(4): 1663–1705.

Brehm, John, and James T Hamilton. 1996. “Noncompliance in environmental

reporting: Are violators ignorant, or evasive, of the law?” American Journal of

Political Science, 444–477.

Burstein, Ariel T, Alexander Monge-Naranjo, et al. 2009. “Foreign Know-How,

Firm Control, and the Income of Developing Countries.” The Quarterly Journal of

Economics, 124(1): 149–195.

Butler, Jeffrey V, Paola Giuliano, and Luigi Guiso. 2016. “The right amount

of trust.” Journal of the European Economic Association, 14(5): 1155–1180.

Calel, Raphael, and Antoine Dechezlepretre. 2016. “Environmental policy and

directed technological change: evidence from the European carbon market.” Review

of Economics and Statistics, 98(1): 173–191.

Cohen, A. 1998. “Monitoring and Enforcement of Environmental Policy.” Interna-

tional Yearbook of Environmental and Resource Economics, 3.

Dasgupta, Susmita, Hemamala Hettige, and David Wheeler. 2000. “What

improves environmental compliance? Evidence from Mexican industry.” Journal of

Environmental Economics and Management, 39(1): 39–66.

Decker, Christopher S, and Christopher R Pope. 2005. “Adherence to environ-

mental law: the strategic complementarities of compliance decisions.” The Quarterly

Review of Economics and Finance, 45(4): 641–661.

Deily, Mary E, and Wayne B Gray. 2007. “Agency structure and firm culture:

OSHA, EPA, and the steel industry.” Journal of Law, Economics, and Organization,

23(3): 685–709.

Duflo, Esther, Michael Greenstone, Rohini Pande, and Nicholas Ryan. 2014.

“The Value of Regulatory Discretion: Estimates from Environmental Inspections in

India.” National Bureau of Economic Research.

24

Dwenger, Nadja, Henrik Kleven, Imran Rasul, and Johannes Rincke. 2016.

“Extrinsic and intrinsic motivations for tax compliance: Evidence from a field exper-

iment in Germany.” American Economic Journal: Economic Policy, 8(3): 203–32.

Ellerman, A Denny, and Barbara K Buchner. 2007. “The European Union emis-

sions trading scheme: origins, allocation, and early results.” Review of Environmental

Economics and Policy, 1(1): 66–87.

European Court of Auditors. 2015. “The integrity and implementation of the EU

ETS.” European Commission.

Evans, Mary F. 2016. “The Clean Air Act Watch List: An Enforcement and Com-

pliance Natural Experiment.” Journal of the Association of Environmental and Re-

source Economists, 3(3): 627–665.

Fehr, Ernst. 2009. “On the Economics and Biology of Trust.” Journal of the European

Economic Association, 235–266.

Fehr, Ernst, and Simon Gachter. 2000. “Cooperation and Punishment in Public

Goods Experiments.” American Economic Review, 90(4): 980–994.

Fehr, Ernst, Urs Fischbacher, Bernhard Von Rosenbladt, Jurgen Schupp,

and Gert G Wagner. 2003. “A nation-wide laboratory: examining trust and trust-

worthiness by integrating behavioral experiments into representative survey.”

Fernandez, R. 2007. “Culture and Economics, New Palgrave Dictionary of Eco-

nomics, edited by Durlauf, S. and Blume, L.”

Fernandez, Raquel, and Alessandra Fogli. 2006. “Fertility: The role of culture

and family experience.” Journal of the European Economic Association, 4(2-3): 552–

561.

Fernandez, Raquel, and Alessandra Fogli. 2009. “Culture: An empirical investi-

gation of beliefs, work, and fertility.” American Economic Journal: Macroeconomics,

1(1): 146–177.

Fisman, Raymond, and Edward Miguel. 2007. “Corruption, norms, and legal en-

forcement: Evidence from diplomatic parking tickets.” Journal of Political Economy,

115(6): 1020–1048.

25

Giuliano, Paola. 2007. “Living arrangements in Western Europe: Does cultural origin

matter?” Journal of the European Economic Association, 5(5): 927–952.

Glaeser, Edward L, David I Laibson, Jose A Scheinkman, and Chris-

tine L Soutter. 2000. “Measuring trust.” The Quarterly Journal of Economics,

115(3): 811–846.

Gray, Wayne B, and Jay P Shimshack. 2011. “The effectiveness of environmen-

tal monitoring and enforcement: A review of the empirical evidence.” Review of

Environmental Economics and Policy, 5(1): 3–24.

Gray, Wayne B, and Mary E Deily. 1996. “Compliance and enforcement: Air

pollution regulation in the US steel industry.” Journal of Environmental Economics

and Management, 31(1): 96–111.

Gray, Wayne B, and Ronald J Shadbegian. 2005. “When and why do plants

comply? Paper mills in the 1980s.” Law & Policy, 27(2): 238–261.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2004. “The role of social

capital in financial development.” The American Economic Review, 94(3): 526–556.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2006. “Does culture affect

economic outcomes?” The Journal of Economic Perspectives, 20(2): 23–48.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2009. “Cultural Biases in

Economic Exchange?” The Quarterly Journal of Economics, 124(3): 1095–1131.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2015a. “Corporate culture,

societal culture, and institutions.” The American Economic Review, 105(5): 336–339.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2015b. “The value of corporate

culture.” Journal of Financial Economics, 117(1): 60–76.

Hallsworth, Michael, John A List, Robert D Metcalfe, and Ivo Vlaev. 2017.

“The behavioralist as tax collector: Using natural field experiments to enhance tax

compliance.” Journal of Public Economics, 148: 14–31.

Jo, Ara, and Stefano Carattini. 2018. “Trust and CO2 Emissions: Cooperation on a

Global Scale.” Grantham Research Institute on Climate Change and the Environment

Working Paper No. 294.

26

Jørgensen, Finn, and Hassa Pedersen. 2005. “Enforcement of speed limitsactual

policy and drivers knowledge.” Accident Analysis & Prevention, 37(1): 53–62.

Luttmer, Erzo F. P., and Monica Singhal. 2014. “Tax Morale.” Journal of Eco-

nomic Perspectives, 28(4): 149–68.

Nyborg, Karine, and Kjetil Telle. 2006. “Firms’ compliance to environmental

regulation: Is there really a paradox?” Environmental and Resource Economics,

35(1): 1–18.

Ostrom, E. 1990. “Governing the Commons: The Evolution of Institutions for Col-

lective Action.”

Ostrom, Elinor. 2000. “Social capital: a fad or a fundamental concept.” Social capital:

A multifaceted perspective, 172(173): 195–98.

Portes, Alejandro. 1998. “Social capital: Its origins and applications in modern

sociology.” Annual Review of Sociology, 24(1): 1–24.

Qin, Botao, and Jason F Shogren. 2015. “Social norms, regulation, and environ-

mental risk.” Economics Letters, 129: 22–24.

Rice, Tom W, and Jan L Feldman. 1997. “Civic culture and democracy from

Europe to America.” The Journal of Politics, 59(04): 1143–1172.

Sapienza, Paola, Anna Toldra-Simats, and Luigi Zingales. 2013. “Understand-

ing trust.” The Economic Journal, 123(573): 1313–1332.

Shimshack, Jay P, and Michael B Ward. 2005. “Regulator reputation, enforce-

ment, and environmental compliance.” Journal of Environmental Economics and

Management, 50(3): 519–540.

Shimshack, Jay P, and Michael B Ward. 2008. “Enforcement and over-

compliance.” Journal of Environmental Economics and Management, 55(1): 90–105.

Tabellini, Guido. 2008. “Presidential address institutions and culture.” Journal of

the European Economic Association, 6(2-3): 255–294.

Tabellini, Guido. 2010. “Culture and institutions: economic development in the

regions of Europe.” Journal of the European Economic Association, 8(4): 677–716.

27

Telle, Kjetil. 2013. “Monitoring and enforcement of environmental regulations:

Lessons from a natural field experiment in Norway.” Journal of Public Economics,

99: 24–34.

Wijen, Frank, and Rob van Tulder. 2011. “Integrating environmental and in-

ternational strategies in a world of regulatory turbulence.” California Management

Review, 53(4): 23–46.

28

Figures and tables for the main text

Figure 1: Correlation between Trust and Noncompliance Rate in the EU ETS

Note: the plot shows a correlation between the level of average trust and compliance rates in the EU

ETS across countries. The level of average trust measure is constructed based on the European Social

Survey (2002-2014).

29

Table 1: Probit Estimation: Trust and Noncompliance in the EU ETSbetween 2005 and 2015

Dependent variable:Indicator for noncompliance

(1) (2) (3) (4) (5)

Trust measured -0.358*** -0.444*** -0.379* -0.468* -1.320**in the country of operation (0.111) (0.150) (0.222) (0.261) (0.657)

Observations 119,701 119,163 119,163 73,498 73,498

Firm-level controls (4) No No No Yes YesCountry-level controls (5) No No Yes Yes YesIndustry FE No Yes Yes Yes YesYear FE No Yes Yes Yes YesClustering Country Country Country Country CountryNumber of clusters 25 25 25 25 25

Notes: The dependent variable in all columns is the binary noncompliance measure thattakes 1 if the installation is out of compliance and 0 otherwise. All estimation is by Pro-bit except for column (5) where I try a logit model. Standard errors are clustered at thecountry of installations’ location. “Industry dummies” are based on the main activity typeinformation provided in the European Transaction Log.

30

Figure 2: Average Noncompliance Rate between 2005 and 2015

Note: the plot shows variation in noncompliance rates across countries. The data on compliance in

the EU ETS is provided by the European Union Transaction Log (EUTL).

31

Figure 3: Average Trust

Note: the plot shows variation in the level of average trust across countries. The level of average trust

measure is based on the European Social Survey (2002-2014).

32

Figure 4: Correlation between Trust in Source Country and Inherited Trust

Note: the plot shows a correlation between inhered trust of second-generation immigrants and the

level of trust in their countries of origin. These measures are constructed based on the European

Social Survey (2002-2014).

33

Table 2: IV Probit Estimation: Trust and Noncompliance in the EU ETSbetween 2005 and 2015

Dependent variable: Indicator for noncompliance

(1) (2) (3) (4) (5) (6)Estimation method IV Probit IV Probit IV Probit IV Probit Probit IV Probit

Trust measured -0.459*** -0.575*** -0.750* -0.865* -0.892***in country of operation (0.118) (0.153) (0.427) (0.461) (0.258)

Inherited Trust -0.494*(0.292)

Observations 119,701 119,163 119,163 73,498 73,498 71,356

Firm-level controls (4) No No No Yes Yes YesCountry-level controls (5) No No Yes Yes Yes YesIndustry FE No Yes Yes Yes Yes YesYear FE No Yes Yes Yes Yes YesClustering Country Country Country Country Country RegionNumber of clusters 25 25 25 25 25 165First stage F stat 40.4 45.0 32.1 60.2 57.7

Notes: The dependent variable in all columns is the binary noncompliance measure that takes 1 if the installation is out of complianceand 0 otherwise. Standard errors are clustered at the level shown in each column. “Industry dummies” are based on the main activitytype information provided in the European Transaction Log. Column (1) shows an IV probit estimate without any controls. Column(2) includes year and industry dummies. Column (3) includes a set of country level controls (GDP per capita, education, populationand two governance indicators). Column (4) further includes several firm-level controls (number of installations each firm owns, totalassets, operating revenue, and number of employees). Column (5) shows the reduced form relationship between noncompliance andthe instrument. Column (6) uses a measure of trust that varies at the region level.

34

Table 3: Trust and Noncompliance in the EU ETS between 2005 and 2015:Exploiting the Differences in the Location of Headquarters

Dependent variable: Indicator for noncompliance(1) (2) (3) (4) (5) (6) (7) (8)

Estimation method Probit Probit Probit Probit Probit IV Probit Probit IV Probit

Trust measured -1.596*** -1.616*** -0.390* -0.417** -0.460* -0.591*** -0.818**in country of central headquarter (0.380) (0.343) (0.204) (0.207) (0.277) (0.222) (0.326)

Trust measured 0.217in region of operation (0.167)

Inherited Trust -0.111***(0.042)

Observations 69,912 51,070 49,174 47,692 33,173 49,160 49,160 33,173

Firm-level controls (4) No Yes Yes Yes Yes Yes Yes YesIndustry FE No No Yes Yes Yes Yes Yes YesYear FE No No Yes Yes Yes Yes Yes YesCountry FE No No Yes Yes Yes Yes Yes YesCountry × year FE No No No No Yes No No YesClustering Country Country Country Country Country Country Country CountryNumber of clusters 25 20 20 20 20 20 20 20

Notes: The dependent variable in all columns is the binary noncompliance measure that takes 1 if the installation is out of compliance and 0otherwise. The sample in this table includes multinational firms only. Standard errors are clustered at the level shown in each column. “Indus-try dummies” are based on the main activity type information provided in the European Transaction Log. Column (1) does not include anycontrols and column (2) adds firm-level controls (number of installations each firm owns, total assets, operating revenue, and number of em-ployees). Column (3) includes year, industry and country of operation fixed effects. Column (4) separately checks the influence of trust in theregion where the installation is located. Column (5) includes country by year fixed effects. Column (6) shows the IV estimate and column (7)presents the reduced form relationship between noncompliance and the instrument. Column (8) further includes country by year fixed effects.

35

Table 4: Trust and Noncompliance in the EU ETS between 2005 and 2015:Robustness Checks

Dependent variable: Indicator for noncompliance(1) (2) (3) (4) (5) (6) (7) (8)

Estimation method IV Probit IV Probit IV Probit IV Probit IV Probit IV Probit IV Probit Probit

Trust measured -0.598*** -0.609** -0.398* -0.602*** -0.615*** -0.599** -1.064*in country of central headquarter (0.223) (0.249) (0.219) (0.222) (0.230) (0.241) (0.562)

Inherited Trust -0.113***(0.044)

Observations 46,257 47,570 49,204 48,921 49,160 49,160 49,160 49,160

Firm-level controls (4) Yes Yes Yes Yes Yes Yes Yes YesRegion-level controls (3) Yes No No No No No No NoIndustry FE Yes Yes Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes Yes Yes YesCountry FE Yes Yes Yes Yes Yes Yes Yes YesClustering Country Country Country Country Country Country Country CountryNumber of clusters 20 18 20 20 20 20 20 20First stage F stat 172.7 191.9 179.8 187.1 194.0 218.7 21.2

Notes: The dependent variable in each column is a binary noncompliance measure that takes 1 if the installation is out of compliance and 0otherwise. Standard errors are clustered at the level shown in each column. “Industry dummies” are based on the main activity type informa-tion provided in the European Transaction Log. Column (1) includes region-level controls and column (2) drops ETS late joiners in my sample(Bulgaria and Romania). In column (3) and (4) I use alternative specifications for the binary compliance variable. In column (5) and (6) I tryalternative measures of trust to check for potential measurement error. In column (7) I try an alternative instrument and column (8) shows thereduced-form relationship between noncompliance and the alternative instrument.

36

Appendix

Bilateral trust

Several papers have looked at the influence of bilateral trust between two countries in

economic activities (Guiso, Sapienza and Zingales, 2009; Bloom, Sadun and Van Reenen,

2012b). The measure of bilateral trust used in the literature comes from the following

question in a series of surveys conducted for the European Commission: “I would like

to ask you a question about how much trust you have in people from various countries.

For each, please tell me whether you have a lot of trust, some trust, not very much

trust, or no trust at all.” This question allows me explore the role of trust in com-

pliance controlling for country of operation and country of origin fixed effects at the

same time and thus solely exploit the pairwise variation in trust. In other words, it will

reveal, for instance, if a French affiliate is more likely to comply with the regulation in

Belgium (that the French tend to trust) than in the United Kingdom (that the French

tend to distrust).

[Table A3]

Column (1) in Table A1 shows that bilateral trust does not play a significant role

in multinationals’ compliance behavior when I include a full set of country of location

and origin dummies (therefore this specification only includes foreign MNEs). The

result is similar when I add firm-level financial variables in column (2). In case the

bilateral trust variable is affected by endogeneity (for instance, better compliance be-

havior in the country of operation might engender trust towards the source country or

there might be omitted bilateral factors affecting trust and compliance behavior at the

same time), I use a measure of religious similarity between two countries developed in

Guiso, Sapienza and Zingales (2009) to instrument for bilateral trust. This measure

is positively correlated with bilateral trust due to long-standing cultural affinities, but

unlikely to affect regulatory compliance exhibited by firms. It yields a strong first

stage (with F statistics of 30) as in previous studies, but the bilateral variable is still

insignificant as shown in column (3). I add an additional instrument in column (4)

that measures somatic distances, based on the average frequency of specific traits (hair

color, height, etc.) present in the indigenous population as in Guiso, Sapienza and

Zingales (2009) since people tend to trust other people who look like them more. The

37

first stage F statistics continues to be strong (around 38). However, the result remains

qualitatively similar.32

This finding can be interpreted in light of the literature on the “transportation” of

culture by individuals across countries. For example, Fisman and Miguel (2007) show

that diplomats from different countries stationed in the same city (the New York City)

display significantly different patterns of corruptive behavior (measured by unpaid

parking fines) that can be explained by the level of corruption in their home countries.

The weak role of bilateral trust together with the strong influence of trust in the

source country is consistent with their findings since it implies that firms operate with

their source-country culture regardless of where they operate.33 Also, another possible

interpretation is related to the importance of knowledge in compliance. Brehm and

Hamilton (1996) and Cohen (1998) distinguish between two sources of noncompliance,

namely, ignorance and evasion. The fact that trust has a strong influence over firms’

compliance but bilateral trust does not, might imply that firms in high-trust countries

may obtain information required for compliance more actively encouraged by the high

anticipated compliance rates by neighboring firms. However, once equipped with the

set of knowledge necessary to comply with the regulation firms may comply wherever

they operate, which makes the role of bilateral trust minimal.

32When I only include the measure of somatic distances as an instrument, the first stage is slightlyweaker with F -statistics 9.7. The result is still qualitatively similar with a coefficient (standard error)of -0.159 (1.073) on the bilateral trust measure.

33Relatedly, it is likely that firms take into account social sanctions coming from their countriesof origin, not from the countries of operation. This interpretation also explains the non-differentialcompliance behavior of firms headquartered in the same country but operate in different countries(thus no effect of bilateral trust).

38

Figure A1: Permit price (spot) between 2005 and 2015

Note: the graph plots the evolution of the EU ETS permit (EUA) price in the spot market. The data

comes from the European Energy Exchange.

39

Table A1: Incidence of Penalties Imposed in the EU ETS Phase 1 by Country

Country 2005 2006 2007Austria No No NoBelgium No No NoBulgaria No

Czech Republic . No NoDenmark No No .Estonia No No NoFinland No No NoFrance No No No

Germany locked locked YesGreece . . No

Hungary No No YesIreland No . .Italy No No No

Latvia No No NoLithuania . . No

Netherlands No No NoNorwayPoland . locked No

Portugal Yes No lockedRomania .Slovakia No No NoSlovenia No No No

Spain Yes Yes YesSweden No No Yes

UK Yes Yes No

Note: ‘.’ indicates that the country did notsubmit the report. ‘locked’ indicates that thecountry submitted the report but access was re-stricted. Bulgaria and Romania joined the EUETS in 2007 and Norway in 2008, thus did nothave reporting obligations in earlier years. Thetable presents the incidence of penalties admin-istered to punish any infringements of the EUETS regulation such as operating without per-mits and not reporting changes in the capacity,as well as excessive emissions.Source: European Environmental Agency Re-porting Obligation Database.

40

Table A2: Distribution of Compliance Code, installation by year observations

Code Frequency PercentA 122,647 93.93B 4,010 3.07C 3,273 2.51D 86 0.07E 563 0.43

Total 130,579 100.0

Source: European Union Transaction Log(EUTL).

41

Figure A2: Noncompliance Rate by Year

Note: the plot shows variation in noncompliance rates across years. The data on compliance in the

ETS is provided by the European Union Transaction Log (EUTL) at the installations level and I

collapse the data over countries to calculate average yearly compliance rates.

42

Table A3: Bilateral Trust and Noncompliance in the EU ETSbetween 2005 and 2015

Dependent variable:Indicator for noncompliance

(1) (2) (3) (4)Estimation method Probit Probit IV Probit IV Probit

Bilateral trust -0.007 -0.059 1.717 0.623(0.254) (0.260) (1.536) (0.804)

Observations 12,292 9,199 7,696 7,696

Firm-level controls (4) No Yes Yes YesCountry of operation FE Yes Yes Yes YesCountry of HQ FE Yes Yes Yes YesYear FE Yes Yes Yes YesIndustry FE Yes Yes Yes YesClustering Country pair Country pair Country pair Country pairNumber of clusters 125 111 77 77First stage F stat 31.5 38.3

Notes: The dependent variable in all columns is the binary noncompliance measure that takes1 if the installation is out of compliance and 0 otherwise. Standard errors are clustered at thecountry of headquarter by country of operation (country pair) level. “Industry dummies” arebased on the main activity type information provided in the European Transaction Log. I usereligious similarity as an instrument in column (3) and somatic distances as well as religioussimilarity in colummn (4). * significant at 10%, ** significant at 5%, *** significant at 1%.

43

Supplementary material for online

publication only (Online Appendix)

Figure OA1: Noncompliance Rate between 2006 and 2015 (Excluding 2005)

Note: the plot shows variation in noncompliance rates across countries, excluding 2005. The data on

compliance in the ETS is provided by the European Union Transaction Log (EUTL) at the installations

level and I collapse the data over all years (excluding 2005) to calculate average compliance rates.

44

Table OA1: Descriptive Statistics: Firm-level Variables

Country Number of Number of Number of Total assets Operating revenuefirms employees installations Thousand USD Thousand USD

per firmAustria 146 712 1.72 626,484 487,542Belgium 248 945 1.54 3,320,201 814,570Bulgaria 112 413 1.35 96,637 107,210Czech Republic 278 850 1.58 279,192 223,508Denmark 269 815 1.60 383,084 424,302Estonia 37 184 1.54 144,959 60,214Finland 183 802 3.68 514,466 412,311France 669 5,281 1.85 3,173,589 1,624,696Germany 1,194 2,730 1.92 1,719,551 1,614,392Greece 104 584 1.46 508,478 409,479Hungary 136 881 2.07 407,152 370,139Ireland 113 4,386 1.39 5,439,998 4,342,960Italy 747 1,037 1.68 971,609 707,199Latvia 75 304 1.49 109,923 59,061Lithuania 78 265 1.49 105,722 127,252Netherlands 327 527 1.43 382,333 588,377Norway 87 599 1.56 1,916,191 1,552,504Poland 584 776 1.70 216,743 216,529Portugal 242 188 1.19 163,830 137,704Romania 209 845 1.34 171,379 153,741Slovakia 152 726 1.35 216,850 203,772Slovenia 91 666 1.10 189,347 142,437Spain 957 529 1.34 448,584 364,320Sweden 279 980 3.09 855,421 493,645United Kingdom 839 6,750 1.65 8,159,246 3,469,543Average 335 1,348 1.68 1,267,680 791,675

Notes: The table reports summary statistics of the financial variables of 8,156 firms used in the regres-sions by country. The data comes from Bureau Van Dijk’s Orbis Database.

45

Table OA2: Descriptive Statistics: Number of Second-generation Immigrants fromEach Source Country

Country of origin Number ofsecond-generation immigrants

Austria 333Belgium 153Czech Republic 382Denmark 143Estonia 38Finland 315France 520Germany 1299Greece 169Hungary 347Ireland 233Italy 971Latvia 78Lithuania 80Netherlands 187Norway 128Poland 835Portugal 186Romania 388Slovakia 373Slovenia 44Spain 251Sweden 149United Kingdom 510

Notes: The table reports the number of second-generation immigrants from each country that I use toestimate inherited trust. The data comes from the Eu-ropean Social Survey.

46

Table OA3: Trust and Noncompliance in the EU ETS between 2005 and 2015:Robustness Checks for Country-level Analysis

Dependent variable: Indicator for noncompliance

(1) (2) (3) (4) (5) (6) (7) (8)Estimation method IV Probit IV Probit IV Probit IV Probit IV Probit IV Probit IV Probit Poisson

Trust measured -0.814** -0.923** -0.835* -0.869* -0.871* -0.837* -0.699 -1.505*in the country of operation (0.351) (0.454) (0.451) (0.483) (0.465) (0.453) (0.430) (0.878)

Observations 69,312 71,158 73,563 73,146 73,498 73,498 73,498 77,558

Firm-level controls (4) Yes Yes Yes Yes Yes Yes Yes YesRegion-level controls (3) Yes No No No No No No NoCountry-level controls (5) Yes Yes Yes Yes Yes Yes Yes YesIndustry FE Yes Yes Yes Yes Yes Yes Yes YesYear FE Yes Yes Yes Yes Yes Yes Yes YesClustering Country Country Country Country Country Country Country CountryNumber of clusters 25 23 25 25 25 25 25 25First stage F stat 35.3 55.7 60.2 61.0 61.8 70.4 37.5

Notes: The dependent variable in each column is a binary noncompliance measure that takes 1 if the installation is out of complianceand 0 otherwise. Standard errors are clustered at the level shown in each column. “Industry dummies” are based on the main activ-ity type information provided in the European Transaction Log. Column (1) adds region-level controls and column (2) drops ETS latejointers (Bulgaria and Romania) in my sample. Column (3) and (4) use alternative specifications for the binary compliance variable. Incolumn (5) and (6), I try alternative measures of trust to check for potential measurement error (detailed explained in the main text).In column (7), I use an alternative instrument and column (8) tries an alternative measure for noncompliance that measures the amountby which installations are noncompliant.

47

Table OA4: Alternative Measure of Trust: Coefficients on the Country Dummiesfrom the European Social Survey

Dependent variable: TrustCountryIndicator Coefficient SDAustria 0.856*** (0.016)Belgium 0.216*** (0.01)Czech Republic 0.134*** (0.014)Germany Reference categoryDenmark 2.288*** (0.003)Estonia 1.191*** (0.012)Spain 0.453*** (0.018)Finland 1.781*** (0.006)France -0.293*** (0.011)United Kingdom 0.598*** (0.005)Greece -0.772*** (0.03)Hungary -0.185*** (0.014)Ireland 0.744*** (0.006)Italy -0.087*** (0.032)Lithuania 0.558*** (0.026)Latvia -0.095*** (0.045)Netherlands 1.049*** (0.005)Norway 1.823*** (0.006)Poland -0.507 (0.019)Portugal -0.314*** (0.058)Romania -0.129 (0.053)Sweden 1.545*** (0.009)Slovenia -0.318*** (0.02)Slovak Republic -0.175 (0.018)

Observations: 283,181R-squared: 0.16

Notes: I report the point estimates on country dummiesused as an alternative measure of trust across countries incolumn (5) and (6) in Table OA3. The coefficients measurethe level of trust in each country relative to Germany, whichis the omitted reference category. Apart from the countrydummies, the regression also included gender, age, educa-tion and income as well as year dummies (not reported).Standard errors are clustered at the country level.Source: European Social Survey 2002 - 2014.

48

Table OA5: Alternative Measure of Trust: Coefficients on the Country Dummiesfrom the World Value Survey

Dependent variable: Trust

Country CountryIndicator Coefficient SD Indicator Coefficient SD

Australia 0.093*** (0.003) Japan 0.014* (0.008)Brazil -0.299*** (0.009) South Korea -0.100*** (0.006)Bulgaria -0.112*** (0.005) Latvia -0.138*** (0.010)Canada 0.000 (0.01) Mexico -0.160*** (0.006)Switzerland 0.055*** (0.005) Malaysia -0.283*** (0.005)Chile -0.201*** (0.007) Netherlands 0.240*** (0.006)China 0.204*** (0.005) Norway 0.332*** (0.005)Cyprus -0.265*** (0.006) Philippines -0.335*** (0.013)Czech Republic -0.084*** (0.01) Poland -0.164*** (0.002)Germany Reference category Romania -0.220*** (0.002)Estonia -0.054*** (0.006) Russian Federation -0.112*** (0.002)Spain -0.123*** (0.006) Saudi Arabia 0.114*** (0.023)Finland 0.176*** (0.006) Sweden 0.255*** (0.006)France -0.164*** (0.01) Singapore -0.107*** (0.012)Hong Kong 0.081*** (0.005) Slovenia -0.179*** (0.004)Hungary -0.071*** (0.01) Slovakia -0.100*** (0.011)Indonesia 0.069*** (0.008) Turkey -0.260*** (0.01)Israel -0.182*** (0.024) Ukraine -0.085*** (0.003)India -0.08*** (0.007) United Kingdom -0.067*** (0.006)Italy -0.037*** (0.01) United States -0.020*** (0.006)Jordan -0.136*** (0.007) South Africa -0.187*** (0.009)

Observations: 263,695R-squared: 0.109

Notes: I report the point estimates on country dummies used as an alternative measure oftrust across countries in column (6) in Table 4. The coefficients measure the level of trust ineach country relative to Germany, which is the omitted reference category. Apart from thecountry dummies, the regression also included gender, age, education and income as well asyear dummies (not reported). Standard errors are clustered at the country level.Source: World Value Survey 1981 - 2013.

49

Proofs

The presence of the threshold type The existence of a cutoff type θ∗, above

which all types violate and below which all types comply with the regulation, directly

follows from the monotonicity of firms’ actions. Firm i chooses an action bi to maximize

its expected payoff written as:

Eui(bi, θi) = −a(bi − θi)2 − (1− a)(bi −Bj)2 − γ

∑i∈N

bi if bi ≤ L (3)

Eui(bi, θi) = −µ[a(L− θi)2 + (1− a)(L−Bj)2 + φ]

− (1− µ)[a(bi − θi)2 + (1− a)(bi −Bj)2]− γ

∑i∈N

bi if bi > L (4)

where (3) represents the expected payoff of abiding by the law and (4) represents the

expected payoff of violating the law.

The first order conditions are:

bi = min[aθi + (1− a)Bj −γ

2, L] if bi ≤ L (5)

bi = aθi + (1− a)Bj −γ

2if bi > L (6)

Note that both (5) and (6) are nondecreasing in θi and (6) is always greater than

(5). Thus, the only possible violation of the monotonicity property is where the payoff-

maximizing action at θi is smaller than (or equal to) L, while at θ′i < θi the payoff-

maximizing action is greater than L. To rule out this scenario, it suffices to show that

for any blow and bhigh such that blow ≤ L and bhigh > L, Eui(bhigh, θi) − Eui(blow, θi) is

increasing in θi.

From (2) and (3), it follows that

Eui(bhigh, θi)− Eui(blow, θi) = −µ[a(L− θi)2 + (1− a)(L−Bj)2 + φ]

− (1− µ)[a(bhigh − θi)2 + (1− a)(bhigh −Bj)2] + a(blow − θi)2 + (1− a)(blow −Bj)

2

(7)

Differentiating with respect to θi yields:

2aµ(L− bhigh) + 2a(bhigh − blow) (8)

It is straightforward to see that the above expression is positive given the definition

50

of blow and bhigh that are smaller (or equal to) and greater than L, respectively for

any a ∈ [0, 1] and µ ∈ [0, 1]. From this monotonicity property, the existence of the

threshold θ∗ follows. QED

θ∗ as a decreasing function of Bj I characterize the expression for a threshold θ∗

by balancing the costs and benefits of violating the regulation for the threshold firm

at θ∗.

Suppose that firm θ∗ decides to violate the regulation. Then the expected payoff

will be:

− µ[a(L− θ∗)2 + (1− a)(L−Bj)2 + φ]

− (1− µ)[a(aθ∗ + (1− a)Bj −γ

2− θ∗)2 + (1− a)(aθ∗ + (1− a)Bj −

γ

2−Bj)

2]

(9)

Suppose instead that firm θ∗ decides to abide by the regulation.

− a(L− θ∗)2 − (1− a)(L−Bj)2 (10)

The threshold θ∗ is given by setting (11) equal to (12). Differentiating both sides

of the resulting equation with respect to B yields:

a(L− θ) ∂θ∗

∂Bj

+ (1− a)(L−Bj) = −a(1− a)(θ∗ −Bj)(∂θ∗

∂Bj

− 1) (11)

Solving for ∂θ∗

∂Bjand simplifying the expression yields 1 − 1

a, which is negative for any

a ∈ [0, 1).

QED

51

Further data descriptions

Trust

European Social Survey I build trust measures using the European Social Survey

(ESS), a collection of cross-country surveys on the individual beliefs, values and social

norms as well as basic demographic information of respondents such as age, education,

religion and occupation, etc. I pool data from the seven waves collected so far (from

2002 to 2014), which includes all European countries in my sample.

The survey elicits trust of respondents by asking the standard question, “Generally

speaking, would you say that most people can be trusted or that you can’t be too

careful in dealing with people?” Answers are given on a scale of 0 to 10, where 0

implies “You can’t be too careful” and 10 means “Most people can be trusted”. The

frequency of individual responses used to build the trust measure by country and wave

is reported in Table OA6.

World Value Survey In Section 5.2, I exploit the difference in the country where

the regulated installations’ global headquarters are located, which include a number

of non-European countries. For this specification, I rely on the World Value Survey

for the data on trust since its geographic coverage is world-wide, while the ESS covers

Europe only. The WVS allows me to exploit the geographical variation in trust across

44 countries shown in Table OA7.

The WVS measures trust by asking the exact same question that appears in the

ESS, which makes the two measures based on the two surveys reasonably comparable.

The only difference is that the answer to the trust question in the WVS is binary, while

the ESS uses a scale of 0 to 10.

Similarly as with the ESS, I pool together seven successive waves administered so

far (1984-2014) and compute the country level trust by taking the simple average over

all observations available for each country available across all waves. The frequency of

individual responses used to build the trust measure by country and wave is reported

in Table A9.

52

Compliance in the EU ETS

The data on compliance behavior in the EU ETS is provided by the European Union

Transaction Log (EUTL), a system harmonized at the EU level that publishes in-

formation on permit allocation, verified emissions and surrendered allowances at the

installation level. I drop countries with less than 50 installations (Cyprus, Iceland,

Liechtenstein, Malta, and Luxembourg). The number of installations in each country

is reported in Table OA8.

In addition to this detailed information, the EUTL also automatically calculates

the compliance status of each installation. There are five possible codes installations

can be given: (1) A, which implies “the number of allowances and ERUs/CERs surren-

dered by 30 April is greater than or equal to verified emissions.”, (2) B, which implies

“the number of allowances and ERUs/CERs surrendered by 30 April is lower than

verified emissions.”, (3) C, which implies “verified emissions were not entered until 30

April.”, (4) D, which implies “verified emissions were corrected by competent authority

after 30 April of year X. The competent authority of the Member State decided that

the installation is not in compliance for year X-1.”, and (5) E, which implies “verified

emissions were corrected by competent authority after 30 April of year X. The compe-

tent authority of the Member State decided that the installation is in compliance for

year X-1.” CERs refer to Certified Emission Reductions and ERUs refer to Emission

Reduction Units (ERUs) from the Clean Development Mechanism (CDM) and Joint

Implementation (JI) that can be used as permits in the EU ETS.

Country-level controls

Governance indicators I use two governance indicators developed by the World

Bank to control for law enforcement or institutional capacity between 2005 and 2015.

One is a measure of country-wide ‘rule of law’ defined as “perceptions of the extent

to which agents have confidence in and abide by the rules of society, and in particular

the quality of contract enforcement, property rights, the police, and the courts, as

well as the likelihood of crime and violence”. The values range between -0.171 and

2.12 in my sample. The second measure is the perceived regulatory quality defined

as “perceptions of the ability of the government to formulate and implement sound

policies and regulations that permit and promote private sector development”. The

values range between 0.213 and 1.921 in my sample. The data can be accessed at:

53

http://data.worldbank.org/data-catalog/worldwide-governance-indicators.

Economic controls For country-level economic controls, I use GDP per capita

in Euro, the percentage of population with tertiary education and total population

between 2005 and 2015. The data comes from the Eurostat. Descriptive statistics for

all country-level controls is reported in Table OA9.

54

Table OA6: European Social Survey: Number of Respondents

Country Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Wave 6 Wave 7 TotalAustria 2,257 2,256 2,405 2,255 2,259 0 1,795 13,227Belgium 1,899 1,778 1,798 1,760 1,704 1,869 1,769 12,577Bulgaria 0 0 1,400 2,230 2,434 2,260 8,324Czech Republic 1,360 3,026 0 2,018 2,386 2,009 2,148 12,947Denmark 1,506 1,487 1,505 1,610 1,576 1,650 1,502 10,836Estonia 0 1,989 1,517 1,661 1,793 2,380 2,051 11,391Finland 2,000 2,022 1,896 2,195 1,878 2,197 2,087 14,275France 1,503 1,806 1,986 2,073 1,728 1,968 1,917 12,981Germany 2,919 2,870 2,916 2,751 3,031 2,958 3,045 20,490Greece 2,566 2,406 0 2,072 2,715 0 0 9,759Hungary 1,685 1,498 1,518 1,544 1,561 2,014 1,698 11,518Ireland 2,046 2,286 1,800 1,764 2,576 2,628 2,390 15,490Italy 1,207 1,529 0 0 0 960 0 3,696Latvia 0 0 1,960 1,980 0 0 0 3,940Lithuania 0 0 0 2,002 1,677 2,109 2,250 8,038Netherlands 2,364 1,881 1,889 1,778 1,829 1,845 1,919 13,505Norway 2,036 1,760 1,750 1,549 1,548 1,624 1,436 11,703Poland 2,110 1,716 1,721 1,619 1,751 1,898 1,615 12,430Portugal 1,511 2,052 2,222 2,367 2,150 2,151 1,265 13,718Romania 0 0 2,139 2,146 0 0 0 4,285Slovakia 0 1,512 1,766 1,810 1,856 1,847 0 8,791Slovenia 1,519 1,442 1,476 1,286 1,403 1,257 1,224 9,607Spain 1,729 1,663 1,876 2,576 1,885 1,889 1,925 13,543Sweden 1,999 1,948 1,927 1,830 1,497 1,847 1,791 12,839United Kingdom 2,052 1,897 2,394 2,352 2,422 2,286 2,264 15,667

Source: European Social Survey (ESS, 2002-2014).

55

Table OA7: World Value Survey: Number of Respondents

Country Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Wave 6 TotalAustralia 1,228 0 2,048 0 1,421 1,477 6,174Brazil 0 1,782 0 0 1,500 1,486 4,768Bulgaria 0 0 1,072 0 1,001 0 2,073Canada 0 0 0 1,931 2,164 0 4,095Chile 0 1,500 1,000 1,200 1,000 1,000 5,700China 0 1,000 1,500 1,000 1,991 2,300 7,791Cyprus 0 0 0 0 1,050 1,000 2,050Czech Republic 0 924 1,147 0 0 0 2,071Estonia 0 0 1,021 0 0 1,533 2,554Finland 1,003 0 987 0 1,014 0 3,004France 0 0 0 0 1,001 0 1,001Germany 0 0 2,026 0 2,064 2,046 6,136Hong Kong 0 0 0 0 1,252 1,000 2,252Hungary 1,464 0 650 0 1,007 0 3,121India 0 2,500 2,040 2,002 2,001 5,659 14,202Indonesia 0 0 0 1,000 2,015 0 3,015Israel 0 0 0 1,199 0 0 1,199Italy 0 0 0 0 1,012 0 1,012Japan 1,204 1,011 1,054 1,362 1,096 2,443 8,170Jordan 0 0 0 1,223 1,200 1,200 3,623Korea, Republic of 970 1,251 1,249 1,200 1,200 1,200 7,070Latvia 0 0 1,200 0 0 0 1,200Malaysia 0 0 0 0 1,201 1,300 2,501Mexico 1,837 1,531 2,364 1,535 1,560 2,000 10,827Netherlands 0 0 0 0 1,050 1,902 2,952New Zealand 0 0 1,201 0 954 841 2,996Norway 0 0 1,127 0 1,025 0 2,152Pakistan 0 0 733 0 0 0 733Philippines 0 0 1,200 1,200 0 1,200 3,600Poland 0 938 1,153 0 1,000 966 4,057Romania 0 0 1,239 0 1,776 1,503 4,518Russian Federation 0 1,961 2,040 0 2,033 2,500 8,534Saudi Arabia 0 0 0 1,502 0 0 1,502Singapore 0 0 0 1,512 0 1,972 3,484Slovakia 0 466 1,095 0 0 0 1,561Slovenia 0 0 1,007 0 1,037 1,069 3,113South Africa 1,596 2,736 2,935 3,000 2,988 3,531 16,786Spain 0 1,510 1,211 1,209 1,200 1,189 6,319Sweden 954 0 1,009 1,015 1,003 1,206 5,187Switzerland 0 1,400 1,212 0 1,241 0 3,853Turkey 0 1,030 1,907 3,401 1,346 1,605 9,289Ukraine 0 0 2,811 0 1,000 1,500 5,311United Kingdom 0 0 1,093 0 1,041 2,134United States 2,325 0 1,542 1,200 1,249 2,232 8,548

Source: World Value Survey (WVS, 1984-2014).

56

Table OA8: Number of Installations in the EU ETS by Country

Country Number of installationsAustria 275Belgium 481Bulgaria 172Czech Republic 464Denmark 455Estonia 65Finland 679France 1,520Germany 2,532Greece 207Hungary 287Ireland 215Italy 1,482Latvia 118Lithuania 124Netherlands 622Norway 173Poland 1,020Portugal 358Romania 284Slovakia 221Slovenia 104Spain 1,362Sweden 875United Kingdom 1,373Total 15,468

Source: European Union Transaction Log(EUTL).

57

Table OA9: Descriptive Statistics: Country-level Variables

Country Rule of Regulatory GDP Tertiary Populationlaw quality per capita education

(Euro) (level 3-8, %)Austria 1.870 1.528 35,645 76.9 8,368,325Belgium 1.360 1.291 33,591 67.5 10,861,533Bulgaria - 0.120 0.600 5,045 73.9 7,429,690Czech Republic 0.962 1.134 14,300 85.5 10,406,087Denmark 1.956 1.821 44,009 70.4 5,530,786Estonia 1.153 1.440 12,273 81.9 1,333,244Finland 1.974 1.769 35,573 76.7 5,352,147France 1.442 1.202 30,909 68.7 64,631,834Germany 1.696 1.571 32,673 79.0 81,483,174Greece 0.593 0.667 18,855 62.5 11,025,804Hungary 0.721 1.023 10,018 75.3 9,990,034Ireland 1.729 1.707 41,555 69.4 4,469,781Italy 0.377 0.840 26,773 53.9 59,184,429Latvia 0.748 1.020 9,882 79.8 2,117,490Lithuania 0.756 1.074 9,918 82.9 3,120,577Netherlands 1.822 1.751 37,791 69.1 16,578,149Norway 1.950 1.495 66,745 74.8 4,868,568Poland 0.620 0.914 9,218 81.8 38,085,752Portugal 1.055 0.926 16,445 34.6 10,512,146Romania 0.014 0.545 6,336 69.5 20,479,399Slovakia 0.503 1.015 11,927 83.4 5,392,052Slovenia 0.962 0.738 17,418 79.5 2,035,400Spain 1.077 1.068 22,845 52.1 45,787,350Sweden 1.930 1.713 40,409 75.8 9,345,354United Kingdom 1.715 1.744 32,909 75.6 62,513,575Average 1.116 1.195 24,355 72.142 19,430,388

Source: Eurostat and the World Bank.

58